[TRTLLM-13629][test] Optimize MoE CI test-db

[xxi-nv]

Summary

Rebalances and trims the MoE-related entries across the per-GPU L0 test-db lists to cut redundant pre-merge cost while preserving (and in places broadening) real-world coverage. Test-list only; no source/runtime changes.

• Stage rebalance (pre -> post-merge): full per-backend matrices (test_moe_backend -k), unquantized (quant=None) and per-tensor FP8 (QDQ) variants of test_moe_module, bf16 4gpus, online/static EPLB, bf16_trtllm_gen, moe_comm_boundary; a representative is kept in pre.
• Dead-entry removal (zero coverage loss): GPT-OSS test_w4_4gpus triton on Blackwell (Hopper-only) and all moe_backend=WIDEEP cases.
• Equivalent-platform dedup (GB200 == B200, SM100): drop B200-side duplicate TestDeepSeekV3Lite 4gpus cases, keep GB200.
• Backend diversification: DeepSeekV3Lite nvfp4_4gpus and the dgx_b200 post set spread moe_backend across CUTLASS/TRTLLM/CUTEDSL per the CutlassFusedMoE SM support table; CUTLASS FP8 buckets split by SM (block-scale in pre, per-tensor QDQ to post).

Test plan

• pre-commit test-list validators pass (entry-existence AST check, waives duplicate check, DCO).
• All 9 modified YAML files parse; no intra-block duplicate entries.
• Refined -k buckets verified against CutlassFusedMoE._QUANT_SUPPORT_TABLE to target only SM-supported quants.

Summary by CodeRabbit

• Tests
  • Refined several integration test matrices across GPU environments.
  • Moved some MoE-related checks to post-merge and narrowed pre-merge coverage for faster, more targeted validation.
  • Updated multi-GPU coverage for DeepSeekV3Lite, GPTOSS, and NemotronV3Super, including adjusted quantization and cache variants.
  • Removed a few legacy or redundant MoE test selections while keeping key backend coverage.

[coderabbitai]

📝 Walkthrough

Walkthrough

Integration test lists are updated to move MoE coverage between pre_merge and post_merge stages, narrow CUTLASS/FP8 selectors, and revise DeepSeekV3Lite, GPTOSS, NemotronV3Super, MoEComm, and legacy fused MoE entries across multiple GPU-specific matrices.

Changes

MoE integration test matrix updates

Layer / File(s)	Summary
H100 stage split tests/integration/test_lists/test-db/l0_h100.yml, tests/integration/test_lists/test-db/l0_dgx_h100.yml	pre_merge MoE entries are narrowed and split from post_merge entries, with CUTLASS and FP8 block-scale cases separated from the remaining CUTLASS variants.
B200/B300 backend relocation tests/integration/test_lists/test-db/l0_b200.yml, tests/integration/test_lists/test-db/l0_b300.yml	Legacy MoE backend entries are removed from pre_merge sections and re-added in post_merge blocks with updated selectors.
DGX B200 selector edits tests/integration/test_lists/test-db/l0_dgx_b200.yml	Nemotron, DeepSeekV3Lite, GPTOSS, and MoE module entries are trimmed and reselected across the pre_merge and post_merge test lists.
DGX B300 and GB200 matrix edits tests/integration/test_lists/test-db/l0_dgx_b300.yml, tests/integration/test_lists/test-db/l0_gb200_multi_gpus.yml	Post_merge DeepSeekV3Lite and NemotronV3Super entries are added, GPTOSS coverage is adjusted, and legacy MoE/MoEComm entries are removed or relocated.
GB300 and RTX cleanup tests/integration/test_lists/test-db/l0_gb300_multi_gpus.yml, tests/integration/test_lists/test-db/l0_rtx_pro_6000.yml	The legacy fused MoE module test is removed from GB300, DeepSeekV3Lite online_eplb variants are narrowed, and commented-out WIDEEP entries are dropped from RTX Pro 6000.

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~60 minutes

Suggested reviewers

• liji-nv
• yuxianq
• hyukn

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description check	✅ Passed	The description covers the summary and testing intent, but it only loosely follows the template and omits an explicit Test Coverage section.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Title check	✅ Passed	The title is concise and accurately reflects the PR’s main change: optimizing the MoE CI test-db.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@tests/integration/test_lists/test-db/l0_b300.yml`:
- Around line 77-80: The l0_b300 test list is unintentionally broader because
the existing `test_moe_backend.py::test_moe_backend -k "CUTEDSL"` and `-k
"DEEPGEMM"` filters also match `MEGAMOE_CUTEDSL` and `MEGAMOE_DEEPGEMM`. Update
this block to keep the MegaMoE variants split out consistently with the other
list, either by adding explicit `MEGAMOE_*` entries in the
`test_moe_backend.py::test_moe_backend` set or by narrowing the `-k` filters so
they do not capture them.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: c83e3041-aa9b-43c3-b649-6d6756d746b8

📥 Commits

Reviewing files that changed from the base of the PR and between a02214a and dcd1f47.

📒 Files selected for processing (9)

• tests/integration/test_lists/test-db/l0_b200.yml
• tests/integration/test_lists/test-db/l0_b300.yml
• tests/integration/test_lists/test-db/l0_dgx_b200.yml
• tests/integration/test_lists/test-db/l0_dgx_b300.yml
• tests/integration/test_lists/test-db/l0_dgx_h100.yml
• tests/integration/test_lists/test-db/l0_gb200_multi_gpus.yml
• tests/integration/test_lists/test-db/l0_gb300_multi_gpus.yml
• tests/integration/test_lists/test-db/l0_h100.yml
• tests/integration/test_lists/test-db/l0_rtx_pro_6000.yml

💤 Files with no reviewable changes (2)

• tests/integration/test_lists/test-db/l0_rtx_pro_6000.yml
• tests/integration/test_lists/test-db/l0_gb300_multi_gpus.yml

[QiJune]

You need to define a new post merge stage like: https://github.com/NVIDIA/TensorRT-LLM/blob/main/jenkins/L0_Test.groovy#L4469.

It seems only 2 case will be moved to post merge, maybe we can revert changes here to avoid defining a new post merge stage.